RaindropA practical guide to choosing decision modelsGitHub ↗

State tracking and updating · question 10

Back to State tracking and updating · All capabilities

Question ID: cruxeval/e9b9076165dab30b931e
Source: CRUXEval
Input format: adapted choices

Explore this question’s altered versions

Context

def f(nums):
    count = len(nums) // 2
    for _ in range(count):
        nums.pop(0)
    return nums

Question

What does f([3, 4, 1, 2, 3]) return?

Answer choices

  • A: [1, 2, 2]
  • B: [1, 2, 4]
  • C: [3, 2, 3]
  • D: [1, 2, 3]

Reference answer: [1, 2, 3] (D)

Exact evaluated input as JSON
{
  "state": "def f(nums):\n    count = len(nums) // 2\n    for _ in range(count):\n        nums.pop(0)\n    return nums",
  "question": "What does f([3, 4, 1, 2, 3]) return?",
  "choices": {
    "A": "[1, 2, 2]",
    "B": "[1, 2, 4]",
    "C": "[3, 2, 3]",
    "D": "[1, 2, 3]"
  }
}

Recorded model answers

Model Answer Outcome
Astra [1, 2, 3] Correct
Opus 5.5 [1, 2, 3] Correct
GLiDE v1.2 [1, 2, 3] Correct
Decider v1.1 [1, 2, 3] Correct
Jev [1, 2, 3] Correct
Microsoft Decision-1 [1, 2, 3] Correct
Clef [1, 2, 3] Correct
Luna Decisions [1, 2, 3] Correct
Clef Flash [1, 2, 3] Correct
Strands Decider [1, 2, 3] Correct
GLiNER2.5 Decide [1, 2, 2] Incorrect
Laya [1, 2, 2] Incorrect
CalDec v1 [1, 2, 2] Incorrect

Source and sharing

Attribution: Meta; CRUXEval authors.
Terms: MIT. Source sharing decision.
The capability assignment and choice-format normalization are part of this experiment. Source data retain their own terms.

Exact source locator and input fingerprint
{
  "source_locator": {
    "path": "data/cruxeval.jsonl",
    "record": "171",
    "kind": "source_file_record"
  },
  "source_revision": "190faf16d175b5847b0af05d937872b1fb395942",
  "input_hash": "d59bdeea565a40467c3959005ab3ce3ecb128b237cefcf53cb1177d4d101950b"
}

Type to search questions, variants and documentation.